Evan Wildenhain

Background

About Evan Wildenhain

In my most recent role at Datasite, I developed solutions using SOTA LLM's mixed with clever data augmentation tricks. I built from the ground up a solution that gathers data and organizes it for other teams usage in models and products. Used many methods such as prompt caching (KV caching on a LLAMA 2 extended with Yarn before prompt caching was publicly offered anywhere), token optimization by swapping out large token lengths with shorter representations or index representations and then supplanting them back into the data after, aswell as token monitoring across multiple Azure Durable function handler instances allowing confidence in the fact that no realtime request would be made without the token bandwith being avaliable to us. The architecture was hosted solely in Azure with an api hosted in Azure webapps, a service bus for splitting and holding messages, and Azure durable functions for the actual functions along with snowflake for final data storage. As a data scientist with a master's degree in computer science from Tufts University, I have a passion for applying machine learning methods to solve real-world problems. I have experience in using PyTorch for data analysis and creating deep neural networks, graph networks, variational autoencoders, and other ML algorithms.